New approach to accident analysis for hazardous road locations

This paper proposes a new approach to the analysis of accidents at hazardous road locations, one that is designed to make these analyses more effective in discriminating between true and false positives, that is, sites whose true expected number of accidents is high and sites where a high recorded number of accidents is predominantly the result of chance variation. The following elements of the proposed approach have not traditionally been part of the analysis of accidents at hazardous road locations: (a) the use of a binomial probability model to estimate the probability that the observed pattern of accidents at a hazardous road location differs from the normal pattern for comparable sites; (b) a two-stage procedure for analysis, in which the first stage consists of an analysis of accidents designed to develop hypotheses about risk factors contributing to the accidents and the second stage is designed to test these hypotheses; (c) the use of a matched-pair, single-blind method for evaluating risk factors that contribute to accidents at hazardous road locations; and (rf) the development of clear criteria for concluding whether an apparently hazardous road location actually is. These elements are designed to guard against some pitfalls of the traditional approach to accident analysis for hazardous road locations, including confirmation of analyst expectancies, data mining, and nonspecific criteria for determining whether a hazardous road location is a true or false positive. The approach is illustrated through simulated data.